How to build a lead scoring model
A lead scoring model is the ruleset that converts raw lead attributes and behavior into a single priority number. The best models are built backward from data: study who actually bought, find the shared signals, and weight them by how strongly they predict a win.
Short answer
Build a lead scoring model by analyzing your closed-won deals to find the attributes and behaviors that predict buying, assigning weighted points to each, setting fit and engagement axes, and defining the threshold for sales handoff. Test it on historical data before going live and recalibrate on real outcomes.
Step by step
Analyze your closed-won deals
Pull your last 6 to 12 months of wins and list the attributes and behaviors they shared before closing. This is the evidence your weights should be based on.
Separate fit signals from behavior signals
Build two axes: fit (industry, size, role) and engagement (actions taken). Keeping them separate prevents a highly active bad-fit lead from scoring high.
Assign weighted points
Give more points to signals that most strongly predicted a win and negative points to disqualifiers.
- Positive: target-industry, decision-maker role, pricing-page visit
- Negative: personal email, competitor, unsubscribed
- Decay: reduce engagement points as activity ages
Set the handoff threshold
Choose the combined score at which a lead becomes sales-ready. Calibrate it so reps get a manageable, high-quality volume, not a flood.
Back-test, launch, and iterate
Run the model against historical leads to confirm high scores map to wins, then launch and refine the weights each quarter as new data lands.
How Ardovo helps
Ardovo derives the model from your own closed-won data rather than guesses, splits fit from engagement, and lets Rook adjust weights automatically as outcomes accumulate. You get a scoring model that improves itself instead of decaying.
Frequently asked questions
Should I build a lead scoring model manually or use AI?
Start with a simple manual model derived from your closed-won data so you understand the logic. As volume grows, let a system recalibrate weights automatically from outcomes. Blind AI scoring you cannot explain is hard to trust or coach against.
What is score decay?
Score decay reduces engagement points over time, so a lead active last week outranks one active six months ago. It keeps the model focused on current intent instead of stale history.
How often should I rebuild the model?
Recalibrate weights quarterly and rebuild more substantially when your ICP, product, or market shifts. A model that never changes slowly drifts away from what actually converts.